Software and systems engineering — Software testing — Part 11: Guidelines on the testing of AI-based systems

This document provides an introduction to AI-based systems. These systems are typically complex (e.g. deep neural nets), are sometimes based on big data, can be poorly specified and can be non-deterministic, which creates new challenges and opportunities for testing them. This document explains those characteristics which are specific to AI-based systems and explains the corresponding difficulties of specifying the acceptance criteria for such systems. This document presents the challenges of testing AI-based systems, the main challenge being the test oracle problem, whereby testers find it difficult to determine expected results for testing and therefore whether tests have passed or failed. It covers testing of these systems across the life cycle and gives guidelines on how AI-based systems in general can be tested using black-box approaches and introduces white-box testing specifically for neural networks. It describes options for the test environments and test scenarios used for testing AI-based systems. In this document an AI-based system is a system that includes at least one AI component.

Ingénierie du logiciel et des systèmes — Essais du logiciel — Partie 11: Lignes directrices relatives aux essais portant sur les systèmes dotés d'IA

General Information

Status
Published
Publication Date
26-Nov-2020
Current Stage
9092 - International Standard to be revised
Completion Date
21-Feb-2022
Ref Project

Buy Standard

Technical report
ISO/IEC TR 29119-11:2020 - Software and systems engineering -- Software testing
English language
52 pages
sale 15% off
Preview
sale 15% off
Preview
Draft
ISO/IEC PRF TR 29119-11:Version 24-okt-2020 - Software and systems engineering -- Software testing
English language
52 pages
sale 15% off
Preview
sale 15% off
Preview

Standards Content (Sample)

TECHNICAL ISO/IEC TR
REPORT 29119-11
First edition
2020-11
Software and systems engineering —
Software testing —
Part 11:
Guidelines on the testing of AI-based
systems
Reference number
ISO/IEC TR 29119-11:2020(E)
©
ISO/IEC 2020

---------------------- Page: 1 ----------------------
ISO/IEC TR 29119-11:2020(E)

COPYRIGHT PROTECTED DOCUMENT
© ISO/IEC 2020
All rights reserved. Unless otherwise specified, or required in the context of its implementation, no part of this publication may
be reproduced or utilized otherwise in any form or by any means, electronic or mechanical, including photocopying, or posting
on the internet or an intranet, without prior written permission. Permission can be requested from either ISO at the address
below or ISO’s member body in the country of the requester.
ISO copyright office
CP 401 • Ch. de Blandonnet 8
CH-1214 Vernier, Geneva
Phone: +41 22 749 01 11
Email: copyright@iso.org
Website: www.iso.org
Published in Switzerland
ii © ISO/IEC 2020 – All rights reserved

---------------------- Page: 2 ----------------------
ISO/IEC TR 29119-11:2020(E)

Contents Page
Foreword .v
Introduction .vi
1 Scope . 1
2 Normative references . 1
3 Terms, definitions and abbreviated terms . 1
3.1 Terms and definitions . 1
3.2 Abbreviated terms .10
4 Introduction to AI and testing .11
4.1 Overview of AI and testing .11
4.2 Artificial intelligence (AI) .11
4.2.1 Definition of ‘artificial intelligence’ .11
4.2.2 AI use cases .12
4.2.3 AI usage and market .12
4.2.4 AI technologies .13
4.2.5 AI hardware .15
4.2.6 AI development frameworks .16
4.2.7 Narrow vs general AI . .16
4.3 Testing of AI-based systems .16
4.3.1 The importance of testing for AI-based systems . .16
4.3.2 Safety-related AI-based systems .17
4.3.3 Standardization and AI .17
5 AI system characteristics .19
5.1 AI-specific characteristics .19
5.1.1 General.19
5.1.2 Flexibility and adaptability .20
5.1.3 Autonomy .20
5.1.4 Evolution .21
5.1.5 Bias .21
5.1.6 Complexity .21
5.1.7 Transparency, interpretability and explainability .22
5.1.8 Non-determinism .22
5.2 Aligning AI-based systems with human values .23
5.3 Side-effects .23
5.4 Reward hacking .24
5.5 Specifying ethical requirements for AI-based systems .24
6 Introduction to the testing of AI-based systems .25
6.1 Challenges in testing AI-based systems .25
6.1.1 Introduction to challenges testing AI-based systems .25
6.1.2 System specifications .25
6.1.3 Test input data .25
6.1.4 Self-learning systems .26
6.1.5 Flexibility and adaptability .26
6.1.6 Autonomy .26
6.1.7 Evolution .26
6.1.8 Bias .26
6.1.9 Transparency, interpretability and explainability .27
6.1.10 Complexity .27
6.1.11 Probabilistic and non-deterministic systems .27
6.1.12 The test oracle problem for AI-based systems .27
6.2 Testing AI-based systems across the life cycle .27
6.2.1 General.27
6.2.2 Unit/component testing .28
© ISO/IEC 2020 – All rights reserved iii

---------------------- Page: 3 ----------------------
ISO/IEC TR 29119-11:2020(E)

6.2.3 Integration testing .28
6.2.4 System testing .28
6.2.5 System integration testing .29
6.2.6 Acceptance testing . .29
6.2.7 Maintenance testing .29
7 Testing and QA of ML systems .29
7.1 Introduction to the testing and QA of ML systems .29
7.2 Review of ML workflow .29
7.3 Acceptance criteria .29
7.4 Framework, algorithm/model and hyperparameter selection .30
7.5 Training data quality .30
7.6 Test data quality .30
7.7 Model updates .30
7.8 Adversarial examples and testing .30
7.9 Benchmarks for machine learning .31
8 Black-box testing of AI-based systems .31
8.1 Combinatorial testing .31
8.2 Back-to-back testing .32
8.3 A/B testing .32
8.4 Metamorphic testing .33
8.5 Exploratory testing .34
9 White-box testing of neural networks.34
9.1 Structure of a neural network .34
9.2 Test coverage measures for neural networks .36
9.2.1 Introduction to test coverage levels .36
9.2.2 Neuron coverage .36
9.2.3 Threshold coverage .36
9.2.4 Sign change coverage .36
9.2.5 Value change coverage .36
9.2.6 Sign-sign coverage .36
9.2.7 Layer coverage .37
9.3 Test effectiveness of the white-box measures .37
9.4 White-box testing tools for neural networks .37
10 Test environments for AI-based systems .38
10.1 Test environments for AI-based systems .38
10.2 Test scenario derivation .39
10.3 Regulatory test scenarios and test environments .39
Annex A Machine learning .40
Bibliography .49
iv © ISO/IEC 2020 – All rights reserved

---------------------- Page: 4 ----------------------
ISO/IEC TR 29119-11:2020(E)

Foreword
ISO (the International Organization for Standardization) and IEC (the International Electrotechnical
Commission) form the specialized system for worldwide standardization. National bodies that
are members of ISO or IEC participate in the development of International Standards through
technical committees established by the respective organization to deal with particular fields of
technical activity. ISO and IEC technical committees collaborate in fields of mutual interest. Other
international organizations, governmental and non-governmental, in liaison with ISO and IEC, also
take part in the work.
The procedures used to develop this document and those intended for its further maintenance are
described in the ISO/IEC Directives, Part 1. In particular, the different approval criteria needed for
the different types of document should be noted. This document was drafted in accordance with the
editorial rules of the ISO/IEC Directives, Part 2 (see www .iso .org/ directives).
Attention is drawn to the possibility that some of the elements of this document may be the subject
of patent rights. ISO and IEC shall not be held responsible for identifying any or all such patent
rights. Details of any patent rights identified during the development of the document will be in the
Introduction and/or on the ISO list of patent declarations received (see www .iso .org/ patents) or the IEC
list of patent declarations received (see patents.iec.ch).
Any trade name used in this document is information given for the convenience of users and does not
constitute an endorsement.
For an explanation of the voluntary nature of standards, the meaning of ISO specific terms and
expressions related to conformity assessment, as well as information about ISO's adherence to the
World Trade Organization (WTO) principles in the Technical Barriers to Trade (TBT), see www .iso .org/
iso/ foreword .html.
This document was prepared by Joint Technical Committee ISO/IEC JTC 1, Information technology,
Subcommittee SC 7, Software and systems engineering.
A list of all parts in the ISO/IEC/IEEE 29119 series can be found on the ISO website.
Any feedback or questions on this document should be directed to the user’s national standards body. A
complete listing of these bodies can be found at www .iso .org/ members .html.
© ISO/IEC 2020 – All rights reserved v

---------------------- Page: 5 ----------------------
ISO/IEC TR 29119-11:2020(E)

Introduction
The testing of traditional systems is well-understood, but AI-based systems, which are becoming more
prevalent and critical to our daily lives, introduce new challenges. This document has been created to
introduce AI-based systems and provide guidelines on how they might be tested.
Annex A provides an introduction to machine learning.
This document is primarily provided for those testers who are new to AI-based systems, but it can also
be useful for more experienced testers and other stakeholders working on the development and testing
of AI-based systems.
As a Technical Report, this document contains data of a different kind from that normally published as
an International Standard or Technical Specification, such as data on the “state of the art”.
vi © ISO/IEC 2020 – All rights reserved

---------------------- Page: 6 ----------------------
TECHNICAL REPORT ISO/IEC TR 29119-11:2020(E)
Software and systems engineering — Software testing —
Part 11:
Guidelines on the testing of AI-based systems
1 Scope
This document provides an introduction to AI-based systems. These systems are typically complex
(e.g. deep neural nets), are sometimes based on big data, can be poorly specified and can be non-
deterministic, which creates new challenges and opportunities for testing them.
This document explains those characteristics which are specific to AI-based systems and explains the
corresponding difficulties of specifying the acceptance criteria for such systems.
This document presents the challenges of testing AI-based systems, the main challenge being the test
oracle problem, whereby testers find it difficult to determine expected results for testing and therefore
whether tests have passed or failed. It covers testing of these systems across the life cycle and gives
guidelines on how AI-based systems in general can be tested using black-box approaches and introduces
white-box testing specifically for neural networks. It describes options for the test environments and
test scenarios used for testing AI-based systems.
In this document an AI-based system is a system that includes at least one AI component.
2 Normative references
There are no normative references in this document.
3 Terms, definitions and abbreviated terms
3.1 Terms and definitions
For the purposes of this document, the following terms and definitions apply.
ISO and IEC maintain terminological databases for use in standardization at the following addresses:
— ISO Online browsing platform: available at https:// www .iso .org/ obp
— IEC Electropedia: available at http:// www .electropedia .org/
3.1.1
A/B testing
split-run testing
statistical testing approach that allows testers to determine which of two systems or components
performs better
3.1.2
accuracy
performance metric used to evaluate a classifier (3.1.21), which measures
the proportion of classifications (3.1.20) predictions (3.1.56) that were correct
© ISO/IEC 2020 – All rights reserved 1

---------------------- Page: 7 ----------------------
ISO/IEC TR 29119-11:2020(E)

3.1.3
activation function
transfer function
formula associated with a node in a neural network that determines the
output of the node (activation value (3.1.4)) from the inputs to the neuron
3.1.4
activation value
output of an activation function (3.1.3) of a node in a neural network
3.1.5
adaptability
ability of a system to react to changes in its environment in order to continue meeting both functional
and non-functional requirements
3.1.6
adversarial attack
deliberate use of adversarial examples (3.1.7) to cause a ML model (3.1.46) to fail
Note 1 to entry: Typically targets ML models in the form of a neural network (3.1.48).
3.1.7
adversarial example
input to an ML model (3.1.46) created by applying small perturbations to a working example that results
in the model outputting an incorrect result with high confidence
Note 1 to entry: Typically applies to ML models in the form of a neural network (3.1.48).
3.1.8
adversarial testing
testing approach based on the attempted creation and execution of adversarial examples (3.1.7) to
identify defects in an ML model (3.1.46)
Note 1 to entry: Typically applied to ML models in the form of a neural network (3.1.48).
3.1.9
AI-based system
system including one or more components implementing AI (3.1.13)
3.1.10
AI effect
situation when a previously labelled AI (3.1.13) system is no longer considered to be AI as technology
advances
3.1.11
AI quality metamodel
metamodel intended to ensure the quality of AI-based systems (3.1.9)
Note 1 to entry: This metamodel is defined in detail in DIN SPEC 92001.
3.1.12
algorithm
ML algorithm
algorithm used to create an ML model (3.1.46) from the training data
(3.1.80)
EXAMPLE ML algorithms include linear regression, logistic regression, decision tree (3.1.25), SVM, Naive
Bayes, kNN, K-means and random forest.
2 © ISO/IEC 2020 – All rights reserved

---------------------- Page: 8 ----------------------
ISO/IEC TR 29119-11:2020(E)

3.1.13
artificial intelligence
AI
capability of an engineered system to acquire, process and apply knowledge and skills
3.1.14
autonomous system
system capable of working without human intervention for sustained periods
3.1.15
autonomy
ability of a system to work for sustained periods without human intervention
3.1.16
back-to-back testing
differential testing
approach to testing whereby an alternative version of the system is used as a pseudo-oracle (3.1.59) to
generate expected results for comparison from the same test inputs
EXAMPLE The pseudo-oracle may be a system that already exists, a system developed by an independent
team or a system implemented using a different programming language.
3.1.17
backward propagation
method used in artificial neural networks to determine the weights to be
used on the network connections based on the computed error at the output of the network
Note 1 to entry: It is used to train deep neural networks (3.1.27).
3.1.18
benchmark suite
collection of benchmarks, where a benchmark is a set of tests used to compare the performance of
alternatives
3.1.19
bias
measure of the distance between the predicted value provided by the ML
model (3.1.46) and a desired fair prediction (3.1.56)
3.1.20
classification
machine learning function that predicts the output class for a given input
3.1.21
classifier
ML model (3.1.46) used for classification (3.1.20)
3.1.22
clustering
grouping of a set of objects such that objects in the same group (i.e. a cluster) are more similar to each
other than to those in other clusters
3.1.23
combinatorial testing
black-box test design technique in which test cases are designed to execute specific combinations of
values of several parameters (3.1.53)
EXAMPLE Pairwise testing (3.1.52), all combinations testing, each choice testing, base choice testing.
© ISO/IEC 2020 – All rights reserved 3

---------------------- Page: 9 ----------------------
ISO/IEC TR 29119-11:2020(E)

3.1.24
confusion matrix
table used to describe the performance of a classifier (3.1.21) on a set of test data (3.1.75) for which the
true and false values are known
3.1.25
decision tree
supervised-learning model (3.1.46) for which inference can be represented
by traversing one or more tree-like structures
3.1.26
deep learning
approach to creating rich hierarchical representations through the training of neural networks (3.1.48)
with one or more hidden layers
Note 1 to entry: Deep learning uses multi-layered networks of simple computing units (or “neurons”). In these
neural networks each unit combines a set of input values to produce an output value, which in turn is passed on
to other neurons downstream.
3.1.27
deep neural net
neural network (3.1.48) with more than two layers
3.1.28
deterministic system
system which, given a particular set of inputs and starting state, will always produce the same set of
outputs and final state
3.1.29
distributional shift
dataset shift
distance between the training data (3.1.80) distribution and the desired
data distribution
Note 1 to entry: The effect of distributional shift often increases as the users’ interaction with the system (and so
the desired data distribution) changes over time.
3.1.30
drift
degradation
staleness
changes to ML model (3.1.46) behaviour that occur over time
Note 1 to entry: These changes typically make predictions (3.1.56) less accurate and may require the model to be
re-trained with new data.
3.1.31
explainability
level of understanding how the AI-based system (3.1.9) came up with a given result
3.1.32
exploratory testing
experience-based testing in which the tester spontaneously designs and executes tests based on the
tester's existing relevant knowledge, prior exploration of the test item (including the results of previous
tests), and heuristic "rules of thumb" regarding common software behaviours and types of failure
Note 1 to entry: Exploratory testing hunts for hidden properties (including hidden behaviours) that, while quite
possibly benign by themselves, could interfere with other properties of the software under test, and so constitute
a risk that the software will fail.
4 © ISO/IEC 2020 – All rights reserved

---------------------- Page: 10 ----------------------
ISO/IEC TR 29119-11:2020(E)

3.1.33
F1-score
performance metric used to evaluate a classifier
...

TECHNICAL ISO/IEC TR
REPORT 29119-11
First edition
Software and systems engineering —
Software testing —
Part 11:
Testing of AI-based systems
PROOF/ÉPREUVE
Reference number
ISO/IEC TR 29119-11:2020(E)
©
ISO/IEC 2020

---------------------- Page: 1 ----------------------
ISO/IEC TR 29119-11:2020(E)

COPYRIGHT PROTECTED DOCUMENT
© ISO/IEC 2020
All rights reserved. Unless otherwise specified, or required in the context of its implementation, no part of this publication may
be reproduced or utilized otherwise in any form or by any means, electronic or mechanical, including photocopying, or posting
on the internet or an intranet, without prior written permission. Permission can be requested from either ISO at the address
below or ISO’s member body in the country of the requester.
ISO copyright office
CP 401 • Ch. de Blandonnet 8
CH-1214 Vernier, Geneva
Phone: +41 22 749 01 11
Email: copyright@iso.org
Website: www.iso.org
Published in Switzerland
ii PROOF/ÉPREUVE © ISO/IEC 2020 – All rights reserved

---------------------- Page: 2 ----------------------
ISO/IEC TR 29119-11:2020(E)

Contents Page
Foreword .v
Introduction .vi
1 Scope . 1
2 Normative references . 1
3 Terms, definitions and abbreviated terms . 1
3.1 Terms and definitions . 1
3.2 Abbreviated terms .10
4 Introduction to AI and testing .11
4.1 Overview of AI and testing .11
4.2 Artificial intelligence (AI) .11
4.2.1 Definition of ‘artificial intelligence’ .11
4.2.2 AI use cases .11
4.2.3 AI usage and market .12
4.2.4 AI technologies .12
4.2.5 AI hardware .15
4.2.6 AI development frameworks .15
4.2.7 Narrow vs general AI . .15
4.3 Testing of AI-based systems .16
4.3.1 The importance of testing for AI-based systems . .16
4.3.2 Safety-related AI-based systems .16
4.3.3 Standardization and AI .16
5 AI system characteristics .18
5.1 AI-specific characteristics .18
5.1.1 General.18
5.1.2 Flexibility and adaptability .19
5.1.3 Autonomy .20
5.1.4 Evolution .20
5.1.5 Bias .20
5.1.6 Complexity .21
5.1.7 Transparency, interpretability and explainability .21
5.1.8 Non-determinism .22
5.2 Aligning AI-based systems with human values .22
5.3 Side-effects .22
5.4 Reward hacking .23
5.5 Specifying ethical requirements for AI-based systems .23
6 Introduction to the testing of AI-based systems .24
6.1 Challenges in testing AI-based systems .24
6.1.1 Introduction to challenges testing AI-based systems .24
6.1.2 System specifications .24
6.1.3 Test input data .25
6.1.4 Self-learning systems .25
6.1.5 Flexibility and adaptability .25
6.1.6 Autonomy .25
6.1.7 Evolution .25
6.1.8 Bias .26
6.1.9 Transparency, interpretability and explainability .26
6.1.10 Complexity .26
6.1.11 Probabilistic and non-deterministic systems .26
6.1.12 The test oracle problem for AI-based systems .26
6.2 Testing AI-based systems across the life cycle .27
6.2.1 General.27
6.2.2 Unit/component testing .27
© ISO/IEC 2020 – All rights reserved PROOF/ÉPREUVE iii

---------------------- Page: 3 ----------------------
ISO/IEC TR 29119-11:2020(E)

6.2.3 Integration testing .27
6.2.4 System testing .28
6.2.5 System integration testing .28
6.2.6 Acceptance testing . .28
6.2.7 Maintenance testing .28
7 Testing and QA of ML systems .28
7.1 Introduction to the testing and QA of ML systems .28
7.2 Review of ML workflow .29
7.3 Acceptance criteria .29
7.4 Framework, algorithm/model and hyperparameter selection .29
7.5 Training data quality .29
7.6 Test data quality .29
7.7 Model updates .29
7.8 Adversarial examples and testing .29
7.9 Benchmarks for machine learning .30
8 Black-box testing of AI-based systems .30
8.1 Combinatorial testing .30
8.2 Back-to-back testing .31
8.3 A/B testing .32
8.4 Metamorphic testing .32
8.5 Exploratory testing .33
9 White-box testing of neural networks.33
9.1 Structure of a neural network .33
9.2 Test coverage measures for neural networks .35
9.2.1 Introduction to test coverage levels .35
9.2.2 Neuron coverage .35
9.2.3 Threshold coverage .35
9.2.4 Sign change coverage .36
9.2.5 Value change coverage .36
9.2.6 Sign-sign coverage .36
9.2.7 Layer coverage .36
9.3 Test effectiveness of the white-box measures .36
9.4 White-box testing tools for neural networks .37
10 Test environments for AI-based systems .37
10.1 Test environments for AI-based systems .37
10.2 Test scenario derivation .38
10.3 Regulatory test scenarios and test environments .38
Annex A Machine learning .40
Bibliography .49
iv PROOF/ÉPREUVE © ISO/IEC 2020 – All rights reserved

---------------------- Page: 4 ----------------------
ISO/IEC TR 29119-11:2020(E)

Foreword
ISO (the International Organization for Standardization) and IEC (the International Electrotechnical
Commission) form the specialized system for worldwide standardization. National bodies that
are members of ISO or IEC participate in the development of International Standards through
technical committees established by the respective organization to deal with particular fields of
technical activity. ISO and IEC technical committees collaborate in fields of mutual interest. Other
international organizations, governmental and non-governmental, in liaison with ISO and IEC, also
take part in the work.
The procedures used to develop this document and those intended for its further maintenance are
described in the ISO/IEC Directives, Part 1. In particular, the different approval criteria needed for
the different types of document should be noted. This document was drafted in accordance with the
editorial rules of the ISO/IEC Directives, Part 2 (see www .iso .org/ directives).
Attention is drawn to the possibility that some of the elements of this document may be the subject
of patent rights. ISO and IEC shall not be held responsible for identifying any or all such patent
rights. Details of any patent rights identified during the development of the document will be in the
Introduction and/or on the ISO list of patent declarations received (see www .iso .org/ patents) or the IEC
list of patent declarations received (see patents.iec.ch).
Any trade name used in this document is information given for the convenience of users and does not
constitute an endorsement.
For an explanation of the voluntary nature of standards, the meaning of ISO specific terms and
expressions related to conformity assessment, as well as information about ISO's adherence to the
World Trade Organization (WTO) principles in the Technical Barriers to Trade (TBT), see www .iso .org/
iso/ foreword .html.
This document was prepared by Joint Technical Committee ISO/IEC JTC 1, Information technology,
Subcommittee SC 7, Software and systems engineering.
A list of all parts in the ISO/IEC/IEEE 29119 series can be found on the ISO website.
Any feedback or questions on this document should be directed to the user’s national standards body. A
complete listing of these bodies can be found at www .iso .org/ members .html.
© ISO/IEC 2020 – All rights reserved PROOF/ÉPREUVE v

---------------------- Page: 5 ----------------------
ISO/IEC TR 29119-11:2020(E)

Introduction
The testing of traditional systems is well-understood, but AI-based systems, which are becoming more
prevalent and critical to our daily lives, introduce new challenges. This document has been created to
introduce AI-based systems and provide guidelines on how they might be tested.
Annex A provides an introduction to machine learning.
This document is primarily provided for those testers who are new to AI-based systems, but it can also
be useful for more experienced testers and other stakeholders working on the development and testing
of AI-based systems.
As a Technical Report, this document contains data of a different kind from that normally published as
an International Standard or Technical Specification, such as data on the “state of the art”.
vi PROOF/ÉPREUVE © ISO/IEC 2020 – All rights reserved

---------------------- Page: 6 ----------------------
TECHNICAL REPORT ISO/IEC TR 29119-11:2020(E)
Software and systems engineering — Software testing —
Part 11:
Testing of AI-based systems
1 Scope
This document provides an introduction to AI-based systems. These systems are typically complex
(e.g. deep neural nets), are sometimes based on big data, can be poorly specified and can be non-
deterministic, which creates new challenges and opportunities for testing them.
This document explains those characteristics which are specific to AI-based systems and explains the
corresponding difficulties of specifying the acceptance criteria for such systems.
This document presents the challenges of testing AI-based systems, the main challenge being the test
oracle problem, whereby testers find it difficult to determine expected results for testing and therefore
whether tests have passed or failed. It covers testing of these systems across the life cycle and gives
guidelines on how AI-based systems in general can be tested using black-box approaches and introduces
white-box testing specifically for neural networks. It describes options for the test environments and
test scenarios used for testing AI-based systems.
In this document an AI-based system is a system that includes at least one AI component.
2 Normative references
There are no normative references in this document.
3 Terms, definitions and abbreviated terms
3.1 Terms and definitions
For the purposes of this document, the following terms and definitions apply.
ISO and IEC maintain terminological databases for use in standardization at the following addresses:
— ISO Online browsing platform: available at https:// www .iso .org/ obp
— IEC Electropedia: available at http:// www .electropedia .org/
3.1.1
A/B testing
split-run testing
statistical testing approach that allows testers to determine which of two systems or components
performs better
3.1.2
activation value
output of an activation function (3.1.3) of a node in a neural network
© ISO/IEC 2020 – All rights reserved PROOF/ÉPREUVE 1

---------------------- Page: 7 ----------------------
ISO/IEC TR 29119-11:2020(E)

3.1.3
activation function
transfer function
formula associated with a node in a neural network that determines the
output of the node (activation value (3.1.2)) from the inputs to the neuron
3.1.4
adaptability
ability of a system to react to changes in its environment in order to continue meeting both functional
and non-functional requirements
3.1.5
adversarial attack
deliberate use of adversarial examples (3.1.6) to cause a ML model (3.1.46) to fail
Note 1 to entry: Typically targets ML models in the form of a neural network (3.1.49).
3.1.6
adversarial example
input to an ML model (3.1.46) created by applying small perturbations to a working example that results
in the model outputting an incorrect result with high confidence
Note 1 to entry: Typically applies to ML models in the form of a neural network (3.1.49).
3.1.7
adversarial testing
testing approach based on the attempted creation and execution of adversarial examples (3.1.6) to
identify defects in an ML model (3.1.46)
Note 1 to entry: Typically applied to ML models in the form of a neural network (3.1.49).
3.1.8
AI-based system
system including one or more components implementing AI (3.1.12)
3.1.9
AI effect
situation when a previously labelled AI (3.1.12) system is no longer considered to be AI as technology
advances
3.1.10
AI quality metamodel
metamodel intended to ensure the quality of AI-based systems (3.1.8)
Note 1 to entry: This metamodel is defined in detail in DIN SPEC 92001.
3.1.11
algorithm
ML algorithm
algorithm used to create an ML model (3.1.46) from the training data
(3.1.80)
EXAMPLE ML algorithms include linear regression (3.1.62), logistic regression, decision tree (3.1.25), SVM,
Naive Bayes, kNN, K-means and random forest.
3.1.12
artificial intelligence
AI
capability of an engineered system to acquire, process and apply knowledge and skills
2 PROOF/ÉPREUVE © ISO/IEC 2020 – All rights reserved

---------------------- Page: 8 ----------------------
ISO/IEC TR 29119-11:2020(E)

3.1.13
autonomous system
system capable of working without human intervention for sustained periods
3.1.14
autonomy
ability of a system to work for sustained periods without human intervention
3.1.15
back-to-back testing
differential testing
approach to testing whereby an alternative version of the system is used as a pseudo-oracle (3.1.59) to
generate expected results for comparison from the same test inputs
EXAMPLE The pseudo-oracle may be a system that already exists, a system developed by an independent
team or a system implemented using a different programming language.
3.1.16
backward propagation
method used in artificial neural networks to determine the weights to be
used on the network connections based on the computed error at the output of the network
Note 1 to entry: It is used to train deep neural networks (3.1.27).
3.1.17
benchmark suite
collection of benchmarks, where a benchmark is a set of tests used to compare the performance of
alternatives
3.1.18
bias
measure of the distance between the predicted value provided by the ML
model (3.1.46) and a desired fair prediction (3.1.57)
3.1.19
classification
machine learning function that predicts the output class for a given input
3.1.20
classifier
ML model (3.1.46) used for classification (3.1.19)
3.1.21
clustering
grouping of a set of objects such that objects in the same group (i.e. a cluster) are more similar to each
other than to those in other clusters
3.1.22
combinatorial testing
black-box test design technique in which test cases are designed to execute specific combinations of
values of several parameters (3.1.54)
EXAMPLE Pairwise testing (3.1.53), all combinations testing, each choice testing, base choice testing.
3.1.23
confusion matrix
table used to describe the performance of a classifier (3.1.20) on a set of test data (3.1.75) for which the
true and false values are known
© ISO/IEC 2020 – All rights reserved PROOF/ÉPREUVE 3

---------------------- Page: 9 ----------------------
ISO/IEC TR 29119-11:2020(E)

3.1.24
pre-processing
part of the ML workflow that transforms raw data into a state ready for use
by the ML algorithm (3.1.11) to create the ML model (3.1.46)
Note 1 to entry: Pre-processing can include analysis, normalization, filtering, reformatting, imputation, removal
of outliers and duplicates, and ensuring the completeness of the dataset.
3.1.25
decision tree
supervised-learning model (3.1.46) for which inference can be represented
by traversing one or more tree-like structures
3.1.26
deep learning
approach to creating rich hierarchical representations through the training of neural networks (3.1.49)
with one or more hidden layers
Note 1 to entry: Deep learning uses multi-layered networks of simple computing units (or “neurons”). In these
neural networks each unit combines a set of input values to produce an output value, which in turn is passed on
to other neurons downstream.
3.1.27
deep neural net
neural network (3.1.49) with more than two layers
3.1.28
deterministic system
system which, given a particular set of inputs and starting state, will always produce the same set of
outputs and final state
3.1.29
distributional shift
dataset shift
distance between the training data (3.1.80) distribution and the desired
data distribution
Note 1 to entry: The effect of distributional shift often increases as the users’ interaction with the system (and so
the desired data distribution) changes over time.
3.1.30
drift
degradation
staleness
changes to ML model (3.1.46) behaviour that occur over time
Note 1 to entry: These changes typically make predictions (3.1.57) less accurate and may require the model to be
re-trained with new data.
3.1.31
explainability
level of understanding how the AI-based system (3.1.8) came up with a given result
3.1.32
exploratory testing
experience-based testing in which the tester spontaneously designs and executes tests based on the
tester's existing relevant knowledge, prior exploration of the test item (including the results of previous
tests), and heuristic "rules of thumb" regarding common software behaviours and types of failure
Note 1 to entry: Exploratory testing hunts for hidden properties (including hidden behaviours) that, while quite
possibly benign by themselves, could interfere with other properties of the software under test, and so constitute
...

Questions, Comments and Discussion

Ask us and Technical Secretary will try to provide an answer. You can facilitate discussion about the standard in here.